I want to use the HTK toolbox for digit recognition. so I downloaded it and read the instruction about how to use it! But when I want to use the HSGen
I encounter with this error:
ERROR [+8250] ReadLattice: 23 Arcs unseen and 15 Nodes unseen
ERROR [+3410] GenSentences: ReadLattice failed
FATAL ERROR - Terminating program HSGen
So the steps I did is as follows:
make a grammer, my file content is exactly as it is stored in a file named gram:
$digit = ONE|TWO|THREE|FOUR|FIVE|SIX|SEVEN|EIGHT|NINE|ZERO;
(SENT-START ($digit) SENT-END)
Using HParse
command to make a network for this grammar, its content is(stored in a file named wdnet) :
I=0 W=!NULL
I=1 W=!NULL
I=10 W=SEVEN
I=11 W=EIGHT
I=12 W=NINE
I=13 W=ZERO
I=14 W=SENT-END
I=2 W=SENT-START
I=3 W=ONE
I=4 W=!NULL
I=5 W=TWO
I=6 W=THREE
I=7 W=FOUR
I=8 W=FIVE
I=9 W=SIX
J=0 S=14 E=1
J=1 S=0 E=2
J=10 S=11 E=4
J=11 S=12 E=4
J=12 S=13 E=4
J=13 S=2 E=5
J=14 S=2 E=6
J=15 S=2 E=7
J=16 S=2 E=8
J=17 S=2 E=9
J=18 S=2 E=10
J=19 S=2 E=11
J=2 S=2 E=3
J=20 S=2 E=12
J=21 S=2 E=13
J=22 S=4 E=14
J=3 S=3 E=4
J=4 S=5 E=4
J=5 S=6 E=4
J=6 S=7 E=4
J=7 S=8 E=4
J=8 S=9 E=4
J=9 S=10 E=4
N=15 L=23
VERSION=1.0
I have a word list file named as wlist2, its content is:
EIGHT
FIVE
FOUR
NINE
ONE
SENT-END
SENT-START
SEVEN
SIX
THREE
TWO
ZERO
I use the HDMan
command to build the dictionary file:
HDMan -m -w wlist2 -n monophones -l dlog2 dict2 mbeep2
which mbeep2 is my dictionary that contains the pronounciations for each word in my wlist2. the ouput file is created in the file named dict2.
When using HSGen with this command:
HSGen -l -n 200 wdnet dict2 > testprompets
I got this error:
ERROR [+8250] ReadLattice: 23 Arcs unseen and 15 Nodes unseen
ERROR [+3410] GenSentences: ReadLattice failed
FATAL ERROR - Terminating program HSGen
I search the error number but coudln't find a related solution! Can anyone help?
Thanks.
Not sure what happened with your wdnet file, but it somehow has lines in reverse order and it also misses few lines. You probably need to repeat the HParse command:
The result should look like this: